Status report on L1Calo energy calibration Yuriy Davygora 1 1 - - PowerPoint PPT Presentation
Status report on L1Calo energy calibration Yuriy Davygora 1 1 University of Heidelberg, Kirchhoff Institute for Physics Level 1 Calorimeter Trigger Joint Meeting, Cambridge, 23-25 March 2011 Outline Introduction Calibration status of autumn 2010
1 University of Heidelberg, Kirchhoff Institute for Physics
Level 1 Calorimeter Trigger Joint Meeting, Cambridge, 23-25 March 2011
2 / 24
3 / 24
Aim: calibrate each TT such that 4 ADC count = 1 GeV “raw” calorimeter transverse energy (LUT calibration important, not subject of this talk) Currently default ADC calibration procedure:
◮ Input: pulser runs (taken by LAr/TILE shifters) ◮ Automatic processing in place: calibration algorithms,
data merging (TILE + LAr), D3PD production, etc.
(see also talks by Peter J. W. Faulkner and Juraj Bracinik)
◮ Output: stored permanently on CASTOR ◮ Implementation: validated calibration sent to Damien Prieur (sqlite file)
Physics data analysis → calibration good, but not perfect In December 2010, for the first time, ADC gains computed from physics data were loaded into L1Calo (details later in this talk)
4 / 24
For more details, see dedicated twiki: https://twiki.cern.ch/twiki/bin/view/Atlas/L1CaloCommissioningChanges
◮ Periods A-C: Inital calibration from Jan/Feb 2010 ◮ Period D: Update of FCAL TT → cell mapping ◮ Period E: FCAL fix ◮ Period G: Presampler HV change ◮ Period J1 (HI): New LUT scheme and LUT slopes
(Stephen Hillier, Alan Watson, William Buttinger, etc.)
◮ Period J2 (HI): Final 2010 calibration (gains from physics data)
Still problems in FCAL: Rx to offline mapping possibly wrong (John Morris is working on that)
5 / 24
6 / 24
Source of Rx gain data: dbTree in John Morris’ D3PDs Here, EM overlap region and FCAL are excluded Some features:
◮ Strong η dependence in EMEC ◮ Structure in φ correlates with cable lengths ◮ Some irregular substructures in φ in HAD layer
7 / 24
Input data: Periods G, H, I (Egamma stream, L1Calo D3PDs p272) Event selection: Standard Egamma GRL Definition: ADC energy (GeV) := (maxADC − pedestal) × 0.25 GeV General cuts
◮ Saturation cut: maxADC <= 1000 ◮ Timing cut: maxADC in the middle of the signal ◮ LUT cut: LUT = 0 ◮ Quality cut (HEC only): quality < 5000
Cuts for L1Calo-Calo deviation analysis
◮ Dead and BadCalo channels discarded ◮ Overlap region 1.4 < |η| < 1.5 in EM layer discarded ◮ Energy cut: 2×ADC + 3×Calo > 6 GeV
Binning in Calo energy (GeV) 4, 7, 10, 14, 18, 22, 26, 30, 40, 50, 60
8 / 24
Difference in digital signal shape: → Difference in analog signal shape → Difference in physics and pulser calibration
9 / 24
Features:
◮ ADC calibration stable in Et, deviation from offline < 2% ◮ LUT problems: LUT “droop”, ADC-LUT offset
10 / 24
EMEC 2 Features:
◮ ADC miscalibration of 10—12%: LAr and presampler problem
(e-mail discussions in autumn 2010)
◮ Also ADC “droop” (yet to be understood)
11 / 24
Examples of ADCvsCalo relative deviation histograms in |η| slices:
12 / 24
These results were obtained from analysis of just several runs of period G, JetTauEtMiss stream Only good channels with smooth and noiseless energy correlation contribute to their respective |η| slices Missing bins: No good channels (quality of channels determined visually)
13 / 24
14 / 24
In December, it was decided to apply calibration corrections from physics data not channel-by-channel, but in |η|-slices New gains were calculated from the data shown on slide 13 Also, in December, the new LUT scheme was implemented and new LUT slopes were loaded (see talk by William Buttinger) The new calibration could be checked with HI data Input: Run 169884 (HI) No GRL selection Cuts: Same cuts as pp except for low energy cut: ADC + 2×Calo > 2 GeV Binning in Calo energy (GeV): 3, 5, 7, 9, 12, 15, 20, 30
15 / 24
Features:
◮ ADC calibration better: deviation < 1% for Et 10 GeV ◮ ADC—LUT offset reduced to almost 0 ◮ Still some residual LUT “droop”
16 / 24
Features:
◮ ADC “droop” still there ◮ Around 1—2% undercalibrated now (Et 20 GeV)
17 / 24
18 / 24
Correlation seems to be OK, calibration awful Same in recent pulser runs (see next slide)
19 / 24
Pulser run 177382 (gain 1) Pulser run 177383 (default gain)
20 / 24
Top left: Second ADC bit seems to be broken (also in pulser, see slide 26) Top right, bottom left: Broken channels not marked as dead Bottom right: Strange noise tails of noise
21 / 24
In EM layer, very often four “broken” channels on same MCM Correlation with OTX maps, i.e. offline and not L1Calo problem (see also backup slides 27 and 28)
22 / 24
23 / 24
Conclusions:
◮ For the first time, L1Calo ADC calibration constants were determined
from physics data, however not yet channel-by-channel
◮ EMEC 2 (PS region, 1.5 < |η| < 1.8) calibration was improved,
still some effects not understood
◮ Channel-by-channel analysis done, some channels with anomalous
features found Outlook:
◮ Deviations between pulser and physics data calibration need to be
studied further (e.g. in pulse shape studies1)
◮ A classification for bad channels (cuts on certain features) has to be
established for automatic channel-by-channel analysis
◮ Long-term plan: calibrate L1Calo channel-by-channel with physics data
(whether as pure physics calibration or as correction to pulser calibration)
1Pulse shape studies have been done by William Buttinger to calculate new FIR coefficients and dropbits for LUT (see his talk) 24 / 24
25 / 24
Pulser run 177383
26 / 24
Colour code
◮ 1 (cyan): marked as bad (visual investigation) ◮ 2 (light green): at least one of 4 layers (PS, layers 1—3)
in OTX maps are broken (OTX map: run 167521)
◮ 3 (red): both visually bad and bad OTX
27 / 24
Same colour code as on slide 27, but now for each layer separately
28 / 24